Image by J. K. Holopainen (Tree Physiology, 2011)

Probing the relationship between formaldehydecolumn concentrations and soil moisture using

mixed models and attribution analysis

STRADA Susanna1, FERNANDEZ-MARTINEZ Marcos2, PENUELASJosep2, FILELLA Iolanda2, YANEZ-SERRANO Ana Maria2, STAVRAKOUTrissevgeni3, BAUWENS Maite3, POZZER Andrea1, and GIORGI Filippo1

1ICTP-ESP, Italy, 2UAB-CREAF, Spain, 3BIRA-IABS, Belgium

Contact: sstrada@ictp.it

6th May 2020 - EGU Sharing On Line, Session: BG3.8

Introduction Methodology Results Next steps

Using formaldehyde as a proxy of isoprene emissions

Emissions

VOCVolatileOrganic

CompoundsHCHO

FormaldehydeOxidation

Remote sensedOMI-HCHO

On ∼100-km scale, HCHO is mainly sensitive to isoprene emissions(Marais et al., 2012; Stavrakou et al., 2018)

Among biogenic volatile organic compound (BVOCs) emitted by vegetation,isoprene is the most abundant (Guenther et al., 2012)

c© Authors. All rights reserved.

Introduction Methodology Results Next steps

Isoprene emissions and water stressWhy do we care about this relationship?

NOx

BVOCs

OZONE POLLUTION

ANTHROPOGENICEMISSIONS

Effect on BVOC driverstemperature, precipitation,

solar radiation, soil moisture

Sunny, warm, calm days

Isoprene emissions affect levels of ozone, methane and particulate matter(Pacifico et al., 2009). Climate change may alter isoprene emissions by modifying

the occurrence and intensity of severe stresses that influence plant functioning(Niinemets, 2010).

c© Authors. All rights reserved.

Introduction Methodology Results Next steps

Isoprene emissions and response to abiotic factorsRadiation, temperature and water avaialbility

Radiation and temperature dependence

Zimmer et al. (2000)

Well established

relationship

Peñuelas and Staudt (2010)

in response to drought No. papers reporting BVOC emission increase, decrease or no-change

Under water stress, emissions decrease or increase?

Niinemets (2010)

Water stress dependence

c© Authors. All rights reserved.

Introduction Methodology Results Next steps

Selected observational global datasets

Variable and dataset Period Resolution Reference

OMI-HCHO (vQA4ECV, L3) 2005–2016 0.25◦ De Smedt et al. (2018)

CRU Temperature (v4.03) 1901–20180.50◦ Univ. East Anglia CRU (2020)

CRU Precipitation (v4.03) 1901–2018

Standardised Precipitation-Evapotranspiration1901–2018 0.50◦ Vicente-Serrano et al., 2010

Index (SPEI, v2.6)

GLEAM Root-Zone Soil Moisture (v3.3b) 1978–2018 0.25◦ Martens et al. (2017)

Copernicus Leaf Area Index (LAI) 2005–2017 0.50◦ Verger et al. (2015)

MODIS C6 Aerosol Optical Depth (AOD)2002–2017 0.05◦ Levy et al. (2013)

(L3, Terra and Aqua)

FLUXCOM Latent Heat 2001–2015 0.50◦ Jung et al. (2019)

All data re-mapped to a 2.5◦ horizontal resolution

Period for analysis: 2005–2016 → Availability of OMI-HCHO observations

Color legend: Aquiring Discarded

c© Authors. All rights reserved.

Introduction Methodology Results Next steps

Linear mixed-effects (LME) modelHow to deal with non-independence in the data?

Linear mixed-effects models are a trade-off between these two alternatives that account for both fixed (variation explained by explanatory variables)

and random (not explained by explanatory var.) effects

By aggregating data at the pixel level, the model is less noisy but some information is lost

Outcome

Predictor

By applying a simple linear regression at each pixel, the model is noisy and does not use all information

ALTERNATIVE 1 ALTERNATIVE 2

Time series in each pixelGlobal dataset

Both spatial and temporal levels of dependence in the dataset

c© Authors. All rights reserved.

Introduction Methodology Results Next steps

Linear mixed-effects modelRandom intercept model

YHCHO = fixed part︸ ︷︷ ︸α + β1X1 + · · · + βqXq

+ random part︸ ︷︷ ︸Site-level variability

+ ε︸︷︷︸Residuals

In the fixed-effect part, the contribution of the explanatory variables and their

interactions is accounted and described using linear regression models.

The random intercept model

assumes that the variation around

the intercept is normally

distributed with a certain variance

c© Authors. All rights reserved.

Introduction Methodology Results Next steps

Linear annual trend in formaldehydeOMI-HCHO version QA4ECV, Level 3 (period: 2005–2016)

By applying a LME model, HCHO does not show an overall trend at the global scale,

while robust trends emerge at the regional scale

-120 -60 0 60 120Longitude (degrees)

-60

-45

-30

-15

0

15

30

45

60

Latit

ude

(deg

rees

)

Theil-Sen standardized slope of OMI-HCHO linear trend (2005-2016)

P-val < 0.05

0.30 0.20 0.15 0.10 0.05 0.02 0.02 0.05 0.10 0.15 0.20 0.30Theil-Sen standardized slope (yr 1 )

Median slope (based on a Theil-sen estimator) of Linear Trend, HCHO annual sums (Period 2005-2016)

c© Authors. All rights reserved.

Introduction Methodology Results Next steps

LME model and temporal contribution analysisRegional scale: Australia

Accounting for only climatic drivers,the LME model (blue line) does notreproduce the observed regional trendin HCHO (black line)⇒ Need to add other explanatoryvariables!

When keeping constant thecontribution over years of oneexplanatory variable at-a-time (redline), SPEI and root-zone soil moistureshow an important contribution

-120 -60 0 60 120Longitude (degrees)

-60

-45

-30

-15

0

15

30

45

60

Lati

tud

e (

deg

rees)

Theil-Sen standardized slope of OMI-HCHO linear trend (2005-2016)

P-val < 0.05

0.30 0.20 0.15 0.10 0.05 0.02 0.02 0.05 0.10 0.15 0.20 0.30

Theil-Sen standardized slope (yr 1 )

Median slope (based on a Theil-sen estimator) of Linear Trend, HCHO annual sums (Period 2005-2016)

∆ Temperature ∆ Precipitation

∆ Root-zoneSPEI Soil Moisture

LME (∆ Temp. const.)

HC

HO

(1

01

5 m

ole

c cm

-2)

HC

HO

(1

01

5 m

ole

c cm

-2)

HC

HO

(1

01

5 m

ole

c cm

-2)

HC

HO

(1

01

5 m

ole

c cm

-2)

Obs.

LME

-120 -60 0 60 120Longitude (degrees)

-60

-45

-30

-15

0

15

30

45

60

Lati

tud

e (

deg

rees)

Theil-Sen standardized slope of OMI-HCHO linear trend (2005-2016)

P-val < 0.05

0.30 0.20 0.15 0.10 0.05 0.02 0.02 0.05 0.10 0.15 0.20 0.30

Theil-Sen standardized slope (yr 1 )

Median slope (based on a Theil-sen estimator) of Linear Trend, HCHO annual sums (Period 2005-2016)

∆ Temperature ∆ Precipitation

∆ Root-zoneSPEI Soil Moisture

LME (∆ Temp. const.)

HC

HO

(1

01

5 m

ole

c cm

-2)

HC

HO

(1

01

5 m

ole

c cm

-2)

HC

HO

(1

01

5 m

ole

c cm

-2)

HC

HO

(1

01

5 m

ole

c cm

-2)

Obs.

LME

c© Authors. All rights reserved.

Introduction Methodology Results Next steps

Conclusions and perspectivesFormaldehyde column concentrations show no overall trend at the global scale, butrobust trends at the regional scale

Including only climatic drivers, the linear mixed-effects model explains more than50% of the observed variance of formaldehyde. However, to correctly reproducethe observed trend, some information is still missing (as observed for Australia)

Over Australia, the Standardised Precipitation-Evapotranspiration Index (SPEI)and the root-zone soil moisture show important temporal contributions

Next stepsInclude as explanatory variables:

Leaf Area Index (LAI) to account for trends in biomass;

Burned fraction to account for trends in wildfires, which are an important source offormaldehyde;

Aerosol Optical Depth to account for anthropogenic sources of formaldehyde.

This project has received funding from the European Union’s Horizon 2020 research

and innovation programme under grant agreement No. 791413

c© Authors. All rights reserved

References

De Smedt et al.: Algorithm theoretical baseline for formaldehyde retrievals from S5P TROPOMI and fromthe QA4ECV project. Atmospheric Measurement Techniques, 2018.

Fernandez-Martınez et al.: Global trends in carbon sinks and their relationships with CO2 and temperature.Nature Climate Change, 2019.

Guenther et al.: The Model of Emissions of Gases and Aerosols from Nature version 2.1 (MEGAN2.1): anextended and updated framework for modeling biogenic emissions, Geoscientific Model Development, 2012.

Jung et al.: The FLUXCOM ensemble of global land-atmosphere energy fluxes. Scientific Data, 2019.

Levy et al.: The Collection 6 MODIS aerosol products over land and ocean, Atmospheric MeasurementTechniques, 2013.

Marais et al.: Isoprene emissions in Africa inferred from OMI observations of formaldehyde columns,Atmospheric Chemistry Physics, 2012.

Martens et al.: GLEAM v3: satellite-based land evaporation and root-zone soil moisture, Geoscientific ModelDevelopment, 2017.

Niinemets: Mild versus severe stress and bvocs: thresholds, priming and consequences. Trends Plant Science,2009.

Pacifico et al.: Isoprene emissions and climate, Atmospheric Environment, 2009.

Penuelas and Staudt: BVOCs and global change. Trends Plant Science, 2010.

Stavrakou et al.: Impact of Short-Term Climate Variability on Volatile Organic Compounds EmissionsAssessed Using OMI Satellite Formaldehyde Observations. Geophysical Research Letters, 2018.

Verger et al.: GEOCLIM: A global climatology of LAI, FAPAR, and FCOVER from VEGETATIONobservations for 1999–2010. Remote Sensing of Environment, 2015.

Vicente-Serrano et al.: A Multi-scalar drought index sensitive to global warming: The StandardizedPrecipitation Evapotranspiration Index - SPEI. Journal of Climate, 2010.

University of East Anglia Climatic Research Unit; Harris, I.C.; Jones, P.D.; CRU TS4.03: Climatic ResearchUnit (CRU) Time-Series (TS) version 4.03 of high-resolution gridded data of month-by-month variation inclimate (Jan. 1901 – Dec. 2018). Centre for Environmental Data Analysis, 2020.

Zimmer et al.: Process-based modelling of isoprene emission by oak leaves. Plant, Cell & Environment, 2000.